Warning;
This was an experimental idea to try and improve the ETH 2.0 shuffling performance. Turns out that it's infeasible to use a more permute-index alike approach, because of random access reads of larger data portions being to slow.
Hashing and writing the outputs are not necessarily the only bottlenecks: when you are working with a large pre-computed list of data, you have to deal with slow reads.
MIT, see license file